SPG: Improving Motion Diffusion by Smooth Perturbation Guidance

2025-03-04

AI Summary

To address out-of-distribution perturbations, kinematic incoherence, and physically implausible motions in human motion diffusion models, this paper proposes a training-free test-time guidance method. Our core innovation is the first formulation of temporal motion smoothing as a weak negative guidance model, imposing structure-preserving constraints during the denoising process. Specifically, we introduce a smooth perturbation mechanism that robustly guides generation without requiring model retraining or fine-tuning, while remaining compatible with diverse motion diffusion architectures. Extensive experiments demonstrate consistent improvements in motion fidelity across multiple models and tasks—including motion prediction, interpolation, and generation—yielding显著 enhancements in joint coherence and physical plausibility. The method is conceptually simple, broadly generalizable, and incurs zero training overhead, offering a practical, plug-and-play solution for enhancing realism in diffusion-based human motion synthesis.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Intelligent Robots: Motion and Path Planning

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

This paper presents a test-time guidance method to improve the output quality of the human motion diffusion models without requiring additional training. To have negative guidance, Smooth Perturbation Guidance (SPG) builds a weak model by temporally smoothing the motion in the denoising steps. Compared to model-agnostic methods originating from the image generation field, SPG effectively mitigates out-of-distribution issues when perturbing motion diffusion models. In SPG guidance, the nature of motion structure remains intact. This work conducts a comprehensive analysis across distinct model architectures and tasks. Despite its extremely simple implementation and no need for additional training requirements, SPG consistently enhances motion fidelity. Project page can be found at https://spg-blind.vercel.app/

Problem

• Improves human motion diffusion model output quality
• Mitigates out-of-distribution issues in motion diffusion
• Enhances motion fidelity without additional training

Innovation

• Test-time guidance without additional training
• Smooth Perturbation Guidance (SPG) method
• Mitigates out-of-distribution issues effectively